CatXG/carememx-benchmark-v0

CareMem-X基准测试v0是一个多语言多模态记忆回忆基准测试数据集，用于评估Gemma 4作为记忆助手的能力。数据集包含200个合成记忆片段，1745个记忆项和1187个回忆问题，支持英语、印地语和Hinglish三种语言。任务类型包括回忆、更新、消歧、跨语言、重复问题、噪声捕获、照顾者交接和不确定性等。每个片段包含记忆列表、回忆问题列表、任务类型和语言信息。

CareMem-X Benchmark v0 is a Multilingual Multimodal Memory Recall Benchmark for evaluating Gemma 4 as a memory assistant. The dataset contains 200 synthetic memory episodes, 1745 memory items, and 1187 recall questions, supporting English, Hindi, and Hinglish languages. Task families include recall, update, disambiguation, cross-lingual, repeated-question, noisy-capture, caregiver-handoff, and uncertainty. Each episode contains memories list, recall questions with expected answers, task families, and languages information.